You Built the AI That’s Firing You. Here’s the Silent Cull Underway.

“Three years ago I trained the model. Three years later the model replaced my team.”

That message arrived at 2 a.m. from a former colleague — a mid-size internet company’s R&D manager. In 2021 his boss told him: go all-in on AI, headcount unlimited. Fresh algorithm grads were getting 250,000–300,000 RMB offers. The company was sending pre-admission letters to juniors. Last month, using Cursor, he built in three days what would have taken five developers two weeks. His boss walked in, laid down the efficiency numbers, and said: “Your group needs to slim down.”

We built the god that sacrificed us. That’s not poetry. That’s the spreadsheet.

This isn’t one guy’s bad luck. It’s the structural logic of an industry that spent a decade stockpiling human code labor and now has to clear it.

——

1. The Software Industry Has Been Overcapacity for a Long Time

Let’s state the ugly truth: The vast majority of developers don’t create anything new. They translate business requirements into standard code blocks. The technical stacks are mature, the frameworks are complete, and the work is cargo-culting — assembling components, wiring APIs, stitching pages.

China’s software sector now employs over 8 million people across 38,000 companies. But ask yourself: how many of them are doing genuinely non-replaceable innovation? The brutal answer is a tiny fraction. The rest are producing what I’ll call ‘low-end code capacity’ — the same back-office systems, e-commerce mini-programs, and CRUD apps that have been built a thousand times before.

Low-end capacity is not just abundant. It’s a liability. When capital retreats and companies need to cut costs, these are the people who get laid off first — not because they’re bad, but because they’re interchangeable. Meanwhile, the market is screaming for people who can build operating systems, databases, and industrial software. We have a surplus of vue.js writers and a desert of kernel developers.

The capacity-clearing mechanism for software is different from steel or cement. No furnaces get shut down. No smokestacks come down. The clearing happens in silence — in the project plan that no longer includes a role you used to fill.

——

2. The Self-Destructive Logic: We’re Not Breaking Our Bowls, We’re Upgrading the Dinner Set

Why do developers themselves push AI tools that make their peers redundant? Because the alternative is worse. In a hyper-competitive market, if you don’t use AI to replace some of your team, your competitor will. The team that delivers the same product in half the time with half the people wins the contract.

But here’s the deeper shift: It’s a transfer of the means of production. Code ability used to live in human brains and muscle memory. Companies had to hire people to get that ability. Now that ability is being distilled into model parameters. A company that subscribes to an AI coding service effectively hires an army of tireless, instantly updating digital craftsmen.

We are not replacing jobs. We are replacing human capacity with algorithm capacity. And the algorithm capacity scales at near-zero marginal cost. The old model — hire 10 developers to write 10,000 lines — is being swapped for: hire one elite developer plus 10 AI agents to write 100,000 lines. The numbers don’t lie.

——

3. The Copernican Moment: You Are No Longer the Centre of the Project

For anyone who’s spent more than a decade in software, the hardest part isn’t the layoff. It’s the decentring. You are no longer the core node. You are becoming middleware between the AI and the business logic.

The old career ladder was clear: junior engineer, senior engineer, architect, CTO. That ladder assumed human coding speed was roughly constant. AI broke that assumption. A senior engineer’s output now expands instantly. The result is a flattened, polarized structure: junior roles get eaten because AI writes basic code better; senior roles get hyper-competitive because you no longer need 10 helpers — you need one great brain and a prompt.

The 1+N model is here: 1 super-individual + N AI agents = what used to take a team of 15.

This isn’t “everyone has a crisis.” It’s a violent filter. The top — people who can define architecture, make trade-offs, hold deep domain knowledge — will become astronomically valuable. The middle — people whose value comes from executing standard tasks — is being liquidated.

——

4. The Silent Capacity-Reduction: No Fires, No Protests, Just Missing Lines in the Budget

Traditional industry capacity reduction comes with explosions. Steel mills get dynamited. Assembly lines are shuttered. There’s drama, protests, and political pain.

Software goes quiet. A requirement document lands. The AI generates the framework. The human reviews, tweaks, and submits. The project ships. And the role that used to write that code from scratch? It’s simply… not in the plan. No one announced it. No one argued. It just stopped being necessary.

The ultimate economic logic of AI coding tools is not to make developers happier. It is to clear deadweight headcount with surgical precision.

We used to think AI would come for the physical laborers first. The cruel irony is that we built the intelligence that understands code, and it turned around and ate the builders.

——

5. How to Not Become the Sacrifice

If you’re a traditional software developer, don’t wallow in the betrayal. Don’t cling to the fantasy that AI is only an assistant. The only useful move is to re-route where you sit on the value chain.

Become the person who defines the problem, not the one who solves it. Go deep into vertical domain knowledge — the kind that isn’t on Stack Overflow. Own the ambiguous reasoning, the value trade-offs, the communication with stakeholders. These things cannot yet be formalised into a prompt.

Here’s the test: if your work can be clearly described in a prompt and the result only needs to be ‘good enough’ — it will be eaten. Your moat is the stuff that cannot be scraped from the public internet: experience, intuition, judgement, responsibility.

When code becomes a generated resource, the only scarce thing left is the ability to decide why something should exist. The how is cheap now. The why is everything.

We started this revolution. We don’t get to stop it. But we can choose which side of the clearing we end up on.